An approach to the development of decision support for diagnosing pathology from radiographs

Large inter-observer variation has been reported in oral radiology. Providing observers with a checklist prompting them to focus on specific radiographically visible features with known relevance to the diagnosis has been shown to improve diagnostic accuracy. We intended to investigate whether diagnostic accuracy for bony pathology could be improved by a computerized decision aid. The decision support provided the user with radiographic reference images for the degree of presence of a radiographic feature (such as radiopacity) and structured the use of diagnostic knowledge.

Twenty-seven general dental practitioners participated in the study. They diagnosed seven radiographic cases of bony pathology using a prototype decision aid and seven other cases unaided. For each radiograph the dentists assessed the degree of presence of five radiographically visible features which had been selected from the literature oral radiology because of their use in describing, radiographically visible bony pathology. The dentists' assessments of degree of feature presence, and their diagnoses, were recorded. Histopathology was used as a diagnostic 'gold standard'. The degree of presence of the features in the radiographic images was also assessed by an independently selected panel of expert oral radiologists in a Delphi consensus procedure.

Determination of the degree of presence of three radiographically visible features assessed by the dentists using the prototype decision aid was significantly improved (by up to 12%) compared to the results of their unaided assessments. The diagnostic accuracy of one-third of the dentists was also improved.

The use of reference images can improve dentists' assessments of the degree of feature presence.